[GUI Plugin] Smart Recommendations

[mantonelli]

Hi!

I've been working on a plugin that recommends books from your own library based on metadata.

What it does

You pick a book, click a button, and the plugin finds the most similar titles already sitting in your library. It compares tags, authors, series, publisher, and publication year to score candidates and sort them by relevance.



The side panel updates as you move through the list, so you can browse covers and metadata without leaving the dialog. Double-click any row (or hit View Book) and Calibre jumps straight to that book.

How the algorithm works

It runs in two steps to stay fast even on large libraries:

Step 1 - Pre-filter: collect books that share at least one tag, author, series, or publisher with the selected book, within the same language. This cuts tens of thousands of books down to a few hundred candidates.

Step 2 - Weighted scoring: each candidate gets a score from 0 to 1 using a weighted sum. The weights shift depending on whether the book looks technical or fiction (detected from tags).

Series-aware scoring: books in the same series are always included as candidates. Adjacent volumes score higher than distant ones - the series weight scales linearly with volume proximity, flooring at 50% for a gap of 10+ volumes.

Features

- Works completely offline - no API keys, no accounts
- Series proximity - volume 3 ranks above volume 12 when you're reading volume 4
- Unread filter - optionally hide books you've already read; same-series books always show up regardless
- Metadata quality report - scan your library for books with incomplete metadata and open them directly in Calibre's metadata editor
- Fast indexing with cache - built once per library and reused across sessions; invalidated automatically when your library changes



Metadata quality report

Since the recommendations depend heavily on metadata quality, the plugin includes a built-in report that scores every book in your library (0–100) and lists exactly what's missing.



You can double-click any row to open Calibre's native metadata editor for that book. The quality dialog stays open, so you can work through the list without reopening it.

Settings



Installation

1. Download `recommender-1.4.0.zip` from the [Releases page](https://github.com/mantonelli/calibr...ation/releases).
2. In Calibre: Preferences > Plugins > Load plugin from file.
3. Select the ZIP and confirm.
4. Restart Calibre.

The Recommend Similar button will appear in the main toolbar.

Tips for better results

The quality of recommendations depends directly on metadata quality. Tags are the most important signal:

# Avoid
Tags: read, to-read, favourite, 2024

# Prefer
Tags: Fantasy, Epic, Trilogy, Brandon Sanderson
Tags: Python, Machine Learning, O'Reilly


The first run builds the index — a few seconds for small libraries, a couple of minutes for very large ones. After that, everything loads from cache instantly.


Source & feedback

GitHub: https://github.com/mantonelli/calibr...recommendation

[BetterRed]

Moderator Notice

Please read the Guidelines in particular #9 relating to inline image sizes. Storing images here is best because it avoids delays and outages elsewhere.

And posting inline thumbs is best.

Curious: where is the index kept.

BR